A cost model is an essential tool for the query optimization in databases. Object-oriented databases have been emerging as a new generation of database technology especially for complex applications such as CAD/CAM, CASE, GIS and OIS. Investigating existing cost models for object-oriented databases, we found several shortcomings which resulted in an inaccurate number of disk accesses and the selection of the query processing strategy which may not be an optimal solution. In this thesis, a new cost model is suggested for much higher accuray. For the centralized object-oriented databases, three kinds of cost formulas are described for three different cases: a single class or a single class hierarchy, a forward class traversal, and a backward class traversal. For the distributed object-oriented databases, one kind of cost formula is described with communication costs. Based on this model, the characteristics of global factorization of common query subexpression (CQS) are analyzed. It turns out that the CQS factorization is very useful in the centralized object-oriented databases and the number of instances is a major factor which determines the CQS effects. In the distributed object-oriented databases, the CQS effects depend on the speed of the network. The CQS effect in the LAN environment is better than one in the WAN since the effect gained by the local processing cost is much bigger than one by the communication cost.